Cactus Needle 2 — mattwood.fyi

I'm Matt Wood, and this is For Your Information. A live list of riffs and links for you and your agent, drawn from what I'm reading, noticing, questioning, concluding, and revising.

Cactus Needle 2

Needle 2 is a 45-million-parameter, 14MB open-source language model designed for tool calling, device control, and structured extraction on low-cost edge hardware like microcontrollers, budget phones, and Raspberry Pis. It runs a full session in 28MB of RAM using CQ2-bit compression, achieving competitive performance against much larger small models on mobile device use benchmarks.

Feels like lots is all happening at once with small mobile-friendly models, but this is a space which has been making steady progress for months now. Encouraging.

cactuscompute.com →

Connections

Supports

AMD acquires AI chip startup Taalas to boost inference performance by etching models into silicon AMD etching models into silicon for inference performance aligns with the same push toward edge/on-device AI inference that Needle 2 targets, representing complementary hardware-software approaches to the same problem

Challenges

Ways to think about token pricing Needle 2's 14MB size at near-zero compute cost directly challenges the token pricing economics discussion — at this scale, pricing models based on cloud inference become irrelevant for edge deployment

Related to

Micro-Agent: Beat Frontier Models with Collaboration inside Model API Both explore making capable AI agents available outside of large frontier model APIs — Micro-Agent through collaboration, Needle 2 through extreme model compression for tool-calling on edge hardware
Inkling: Thinking Machines' Open-Weights Model (and Tinker Fine-Tuning Platform) Both are open-weights small language models positioned as alternatives to large frontier models, emphasizing accessibility and local deployment

Type:	link
First published:	Tuesday, 11 August 2026 at 01:10 UTC
Location:	https://cactuscompute.com/needle
URL:	https://mattwood.fyi/i/e4b6aaeb